This document summarizes the 2008 International Workshop on Recommendation Systems for Software Engineering (RSSE). It introduces recommendation systems and their application to software engineering. Key topics discussed include long-standing questions about how recommendation systems work and emerging questions around their use for software development. The goals of the workshop are to identify new themes in the field, results ready for inclusion in recommendation systems, current shortcomings, and future directions for research.
9. 5‐minute madness
• Gail:
– “The recommenda2on conundrum”: what do people need
help with, how to deal with compe2ng recommenders
– recommenda2ons work well when task‐first rather than
tool‐first
• Adrian:
– Interac2on effects with other recommenders, either
synchronously or sequen2al
• Rob:
– announcement of a data mining summer school
(rob.deline@microso.com)
11. Current shortcomings
• Missing data impedes analyses (e.g., method
bindings)
• Too many (false‐posi2ve) recommenda2ons
• Ra2onales are oen absent
– Apparent tradeoff: usefulness/ease of provision
• User‐specific needs and staleness of those needs
• Lack of feedback capture
• Screen real estate needed
12. Themes
• Thoughts on what recommenda2on systems
generally involve
• Features of recommenda2on systems
– recommenda2on filtering
– ranking/ordering
– ra2onale
– means of presen2ng the results
• Cost model
• Level of granularity
14. Stable results
• Recommenda2ons can involve people,
ar2facts, or both
• Dimensions of recommenda2on systems
– popula2on set (what might be recommended)
– possibly, addi2onal data (analyzed to derive
results)
– result set
– ???
• ???
15. Future direc2ons
• Guidelines for what cons2tutes good eval. (rela. to maturity)
– user studies vs. simula2ve studies
• strengths/weaknesses
– benchmarks to avoid user studies, improve comparability
• via Wizard of Oz studies
• real user studies are necessary eventually, though
• What cons2tutes a good ra2onale
– Levels of abstrac2on approach suggested
• Understanding usage modes:
– When is it (not) appropriate to push informa2on, and how
prominently?
– Should systems only provide batch‐mode support or in‐context‐mode
support, or always both
• “Ungameable” recommenda2on systems?
• “Unified theory of RSSE”